Re: [Qemu-devel] [PATCH v2 3/4] raw-posix: Fix try_seek_hole()'s handling of SEEK_DATA failure

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 3829 bytes --]

On 11/13/2014 08:29 AM, Eric Blake wrote:

>>> lseek() with SEEK_DATA starting in a hole when there is no data until
>>> EOF is actually the part that isn't documented in the man page, but
>>> ENXIO is what I'm seeing here on RHEL 7.
>>
>> Here's the (proposed) POSIX wording:
>>
>> http://austingroupbugs.net/view.php?id=415
>>
>> And ENXIO is indeed the expected error for SEEK_DATA on a trailing hole,
>> so maybe we should special case it.
>>
> 
> Uggh.  Historical practice on Solaris (and therefore the POSIX wording)
> says that SEEK_HOLE in a trailing hole is allowed (but not required) to
> seek to EOF instead of reporting the offset requested.  I have no clue
> why this was done, but it is VERY annoying - it means that if you
> provide an offset within a tail hole of a file, you cannot reliably tell
> if the file ends in a hole or with data, without ALSO trying SEEK_DATA.
>  For applications that are reading a file sequentially but skipping over
> holes, this behavior is fine (it short-circuits the hole/data search
> points and might shave an iteration off a lop).  But for OUR purposes,
> where we are merely trying to ascertain whether we are in a hole, we
> have an inaccurate response - since SEEK_HOLE does NOT return the offset
> we passed in, we are prone to treat the offset as belonging to data,
> which is a pessimization (you never get wrong results by treating a hole
> as data and reading it, but it is definitely slower).
> 
> I think you HAVE to call lseek() twice, both with SEEK_HOLE and with
> SEEK_DATA, if you want to accurately determine whether an offset happens
> to live within a trailing hole.

Here's a table of possible situations, based solely on POSIX wording
(and not on actual tests on Solaris or Linux, although it shouldn't be
too hard to confirm behavior):

0-length file:
lseek(fd, 0, SEEK_HOLE) => -1 ENXIO
lseek(fd, 0, SEEK_DATA) => -1 ENXIO
conclusion: 0 is at EOF

file of any size:
lseek(fd, size_or_larger, SEEK_HOLE) => -1 ENXIO
lseek(fd, size_or_larger, SEEK_DATA) => -1 ENXIO
conclusion: size_or_larger is at or beyond EOF

file where offset is in a hole, but data appears later:
lseek(fd, offset, SEEK_HOLE) => offset
lseek(fd, offset, SEEK_DATA) => end_of_hole
conclusion: offset through end_of_hole is in a hole

file where offset is data, whether or not a hole appears later:
lseek(fd, offset, SEEK_HOLE) => end_of_data
lseek(fd, offset, SEEK_DATA) => offset
conclusion: offset through end_of_data is in data

file where offset is in a tail hole, option 1:
lseek(fd, offset, SEEK_HOLE) => offset
lseek(fd, offset, SEEK_DATA) => -1 ENXIO
conclusion: offset through EOF is in hole, but another seek needed to
learn EOF

file where offset is in a tail hole, option 2:
lseek(fd, offset, SEEK_HOLE) => EOF
lseek(fd, offset, SEEK_DATA) => -1 ENXIO
conclusion: offset through EOF is in hole, no additional seek needed

The two calls are both necessary, in order to learn which extant type
offset belongs to, and to tell where that extant ends; and the behaviors
are distinguishable (if both lseek() succeed, we have both numbers we
want; if both fail with ENXIO, we know the offset is at or beyond EOF;
and if only SEEK_HOLE fails with ENXIO, we know we have a trailing
hole); and we can tell at runtime what to do about a trailing hole (if
the return value is offset, we need one more lseek(fd, 0, SEEK_END) to
find EOF; if the return value is larger than offset, we have EOF for
free).  You can optimize by calling SEEK_HOLE first (if it fails with
ENXIO, there is no need to try SEEK_DATA); but SEEK_HOLE in isolation is
insufficient to give you all the information you need.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 539 bytes --]